It is a well-recognized challenge to monitor accurately physical demands on cardiac support imposed by different levels of work or exercise a patient engages in (Barold et al., Clin Cardiol (1997), Vol 20 (8): 726-9; Alt, Am J Cardiol (1999), Vol 83(5B): 17D-23D). And yet, it is essential for a cardiac pacemaker to operate at a pacing rate that correlates with the workload presented. It is equally essential for a cardioverter defibrillator to determine hemodynamic status of a patient before diagnosing a life-threatening arrhythmia and exerting ventricular anti-tachy therapy (Johnston et al., Eur Heart J (1998), Vol. 19 (12):1879-88). A variety of physiologic or non-physiologic sensors have been designed during the last two decades to produce a signal directly related to the metabolic demand, including blood pH (U.S. Pat. No. 4,009,721), QT interval (U.S. Pat. No. 4,228,803), blood O2 saturation (U.S. Pat. No. 4,399,820), blood temperature (U.S. Pat. No. 4,543,954), pressure sensor (U.S. Pat. Nos. 4,899,752 and 5,105,819), piezoelectric crystal (U.S. Pat. Nos. 4,140,132 and 4,428,378), accelerometer (U.S. Pat. Nos. 5,235,237 and 5,383,473), micro-accelerometer (U.S. Pat. Nos. 5,423,883, 5,480,412, Rickards et al., The Multicenter PEA Study Group—PACE (1996) 19:12 Pt1, 2066-2071), thoracic and intracardiac impedance (U.S. Pat. Nos. 3,593,718, 4,291,699, 4,773,401, 5,085,583, 5,235,976, 5,562,711 and 5,782,884) etc. These sensors are less than optimal and have led to little commercial success for various reasons, including difficulty to implement due to the need for specific hardware, lack of correlation between the signal and actual workload, lack of hemodynamic feedback information, slow response, and lack of robustness, i.e., high sensitivity to noise or artifacts inherent to the signal.